Weight partitioned Probability Hypothesis Density filters

The Probability Hypothesis Density filter gives an estimate of the multistate solution set without a multidimensional assignment between measurements and the target estimates. The filter itself outputs a multimodal surface from which individual target estimates must be manually extracted. Furthermore, since the filter propagates the entire multistate estimate, it does not provide any natural connection between any individual state estimates extracted at consecutive timesteps. Recently a new series of deconvolution methods known as CLEAN algorithms have been explored in the particle-based PHD context as a new method of state extraction which considers both the weight and spatial properties of the state estimates. This paper explores weight based state extraction in PHD filters in a more general context and focuses on the issue of track continuity when using weight partitioned PHD filters. The partitions are maintained over time, based on their spatial and weight characteristics so to represent individual or singleton estimates at each timestep.